ChatGPT helps writers with many tasks such as conducting research, expanding on their ideas, bouncing content strategies, and generating rough drafts. Individual authors and content teams use it in different ways to enhance their writing workflows.
While there is no perfect prompt, there are certain principles that guide how you prompt an LLM to produce useful text. These principles depend on:
Your objectives and goals for the content you are creating
The fundamentals of content writing and marketing
Resources available (time and money, because tools are expensive)
The limitations of AI tools themselves
Each of the principles from each category can be translated into effective actions, which we term “best practices.”
In this article, I have listed five such best practices that come from the principles related to the limitations of AI tools.
Let’s look at what needs to be done, how you should do it, the expected advantages, and the principle that started it all.
1. Generate the material you need
GenAI is great at collecting information from multiple sources and presenting it in a particular context. Consequently, it is a phenomenal learning and research tool. Unfortunately, this also makes LLMs pretty shit at writing persuasive content.
There are plenty of writers who straight up publish AI-generated content but they are hurting their long-term credibility and profitability. A collection of information in a particular sequence will fail to aptly engage the reader. It lacks the warmth and compassion you need to convey as a writer to earn your audience’s trust and motivate them to take a certain action.
That can be a bit disappointing for those who started this course, but it’s actually for the better.
As I mentioned above, AI is great at collecting information from multiple sources and presenting it in a particular format, style, context, etc., it speeds up your research process. The research process is actually the most time-consuming part of content writing.
A GPT-generated draft, for instance, contains all the information you need in the format of the article you are about to write. Of course, you need to do a tad bit of work in finding credible sources of information, but the machine will extract the relevant points and organize them in minutes.
The previous chapter of this book (linked below) outlines the workflow and prompts will help you do that effectively. I’ve also linked my custom GPTs that make the entire workflow user-friendly.
2. Just give enough details
When I just started experimenting with different LLMs, my prompts were too long and complex. I included unnecessary details such as the priming sentences “Imagine you are an experienced personal trainer” and broader objectives “I wish to inspire my readers to be fit.”
The “AI Gurus” across the internet are still suggesting it’s a great idea to bloat your command to ChatGPT and make the subjective claim of “improved results.” However, the opposite is true.
Just tell it what you want and the format of the output. For example:
“Give me 10 advantages of working out in the morning for beginners. Keep the bullets short one-liners.”
If you add irrelevant things like the ones mentioned earlier, you are fundamentally tanking your productivity. The degree of improvement in the output is intangible and extremely subjective. Even if there are any improvements, it’s too minimal to be noticeable.
Plus, additional data increases the chance of hallucination. The LLM has to consider a lot more and a lot of diverse sets of tokens while generating an output. This is the biggest time-waster, particularly if you spend minutes writing large prompts.
3. Verify sources manually
Consider you are generating a long article draft with ChatGPT, one section (subheading) at a time. Let’s say there are some sections that discuss the latest and most relevant information on the topic of the article.
Doing it manually will beat the purpose of using AI because it will cost you almost the same time.
Here’s what you do instead — mention the source web pages in the outline. When you feed it to ChatGPT, one section at a time, it will visit the URL, extract the relevant data, and include it while generating content.
A common mistake you may be enticed to make is to ask ChatGPT to use the internet and conduct autonomous research. I tried it too.
The truth is, ChatGPT used Bing because it’s owned by Microsoft, and the possibility of it sourcing information from an irrelevant web page is too high for my liking.
Better use Perplexity.ai and authenticate the web page yourself. It is faster and there are little chance of ChatGPT getting it wrong. But be careful, GPT-4o cannot read dynamic pages, so you may have to find a workaround for that.
There are a couple of strategies for dynamic pages, actually. One, get the info manually and mention it in the outline. Two, take a full-page screenshot and upload it while generating that particular section. Test them both and see what works best for you.
4. Do the creative work
An idea for an article begins with a statement, supported by an argument. AI helps you expand on both and collect the relevant information, making it easier for you to convert the idea into a full-fledged article.
When you make an argument, it also needs to be engaging and persuasive. The problem is these two are intangible variables that remain unique to each article.
Fortunately, this is the best part of content writing — creativity.
Creativity in content writing refers to two things:
The logical pathway chosen by the author to justify their statement.
The emotions that the author chooses to evoke along the way.
Neither can’t be done by AI, so stop trying to do that. Besides, if you don’t enjoy the creative part of crafting an article or any kind of content, consider a career change. There is enough shit content on the internet, we don’t want more.
5. Experiment with new prompts
Prompt engineering 101.
Initially, your prompts will be too long. Then, you will shorten them like I did. Then, you will probably create custom GPTs that trim your prompts even further when writing content.
Eventually, you would prefer writing the prompts on the fly.
Tweak your prompts here and there and try them out on multiple LLMs. Compare the results to determine which set of prompts and tools deserve a place in your workflow. It’s a good practice to journal your progress as it will make it simpler for you to store those prompts along the way.
To level your prompting game up even further, try writing different kinds of content. Humor, political, spiritual, and emotional content. This will help you dive into different aspects of a piece and allow you to elevate your game as a writer too.
Whatever you do, NEVER purchase any kind of prompting course or pay for a king of prompting playbook. These ones are often trash and the best prompts are discovered by practice.
Conclusion: Just start now
The first two chapters of this book were to introduce you to the entire workflow and give you some best practices. The next step for you is to start putting it to work.
Initially, you will be slower in your output. Figuring out the steps for yourself and going back and forth between my chapters and your workspace will add to the inefficiency. Furthermore, you need to double-check everything as it is new.
If you write one article with it for five days, you can start seeing significant gains on day six. After you write an article with ChatGPT, analyze what you did in your mind to discover areas of improvement.
The rest of the chapters of this book dive into the nuances of AI, writing, and marketing to refine and polish your thought process around the GPT-powered workflow.
—
Next Chapter: Generative AI in Content Writing: Pros and Cons
Previous Chapter: The Complete ChatGPT Content Writing Workflow
Index (with Prologue): Content Writing With AI: Principles, Strategies, and Prompts
Comments